Three Essays in Finance

Abstract

Dynamic factor models for forecasting stock return covariance matrix. Factor models are quite popular for forecasting covariances as they reduce the impact of the idiosyncratic return component on forecasts based on the full sample covariance matrix. I examine empirically whether introducing factor dynamics brings any benefits in the context of forecasting the return covariance matrix. When dynamic factor models are used in a global minimum variance problem, the resulting portfolio has marginally lower volatility than when naive historical or static factor models are used. Slightly sharper results in favor of the dynamic factor models are obtained when looking for the minimum tracking error portfolio and when estimating VaR. However, the differences between dynamic and static factor models are not overwhelming.